e2e

Teams building software with coding agents have sped up development, but verification has not kept pace. Classic end-to-end tests use fixed locators and break on every small UI change, while handing testing entirely to an AI agent means burning tokens re-running the same clicks on every CI run. Developers quoted on the page describe being stuck choosing between brittle deterministic suites and flaky, expensive pure-agent testing, and say they were already trying to build a hybrid solution themselves.

e2e lets developers write tests as plain TypeScript files in their own repo that combine precise deterministic locators and assertions with natural-language agent actions (for example 'add two items to the cart, then go to checkout'). It runs via npx in the terminal or CI, works across web, iOS and Android from day one, and is model agnostic, using the user's own API keys or subscriptions rather than a hosted model. A key feature is caching: once an agent step succeeds, the run is replayed on future tests without calling the model again, which the page and multiple users cite as the way it cuts token cost and reduces flakiness compared to pure-agent testing.

The page frames itself around the shift to agentic coding: 'building software got fast. verification did not,' arguing that as AI agents write more code, a verification layer built for that workflow is needed.